Estimating averages by groups for a matched sample of companies

[Nadya Mihaylova]

Hello,

I would like to calculate the average turnover of treated companies and compare the value to a matched sample of non-treated companies. The matched pair IDs are shown in a variable called "pairs".

Can you please help me estimate the values below using a small subset of my dataset as an example?

average turnover of treated companies = (132 + 27)/2
average turnover of matched non-treated companies = (128 + 26)/2

Thank you.

company	year	treated	turnover	pairs
Company A	2014	0	134	1
Company A	2015	1	132	2
Company A	2016	0	145	3
Company B	2014	0	129	1
Company B	2015	0	128	2
Company B	2016	0	122	3
Company C	2014	0	23	4
Company C	2015	0	24	5
Company C	2016	1	27	6
Company D	2014	0	20	4
Company D	2015	0	21	5
Company D	2016	0	26	6

[Carlo Lazzaro]

Nadya:
you may want to try:

. bysort treated: sum turnover

---------------------------------------------------------------------------------------------------------------------------
-> treated = 0

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
    turnover |         10        77.2    57.65183         20        145

---------------------------------------------------------------------------------------------------------------------------
-> treated = 1

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
    turnover |          2        79.5    74.24621         27        132


.

[Nadya Mihaylova]

Hi Carlo,

Thank you for your response. I thought of doing it this way, but the average turnover for no-treated businesses is based on 10 observations (rather than 2 as in my example). I want the estimate to be equal to 77.0 and not 77.2.

In other words, I only want to include matched companies.

[Carlo Lazzaro]

Nadya:
sorry, but I'm not clear with what you're after.

[Nadya Mihaylova]

Ok, I will try to explain it.

I used Propensity Score Matching to match treated companies with non-treated companies based on observable business characteristics (such as industry, location, age etc). Matched pairs are given pair IDs (saved in a new variable called pairs). So, in 2014, company A is matched with company B.

I would like to estimate the average turnover of all treated businesses and compare that to non-treated business with the same pair IDs. So, if my average for the treatment group is based on company A (2015) and Company C (2016), I need the average for the non-treated group to be based on Company B (2015) and company D (2016).

[Øyvind Snilsberg]

perhaps,

bys pairs: egen matched = max(treated==1)
bys treated: sum turnover if matched